// Camera / view operations on `Scene`: zoom, fit-all, named-view restore,
// camera<->document sync, and per-tile grid/snap. (The actual screen-space
// hit-testing lives in `scene::pick::hit_test`.)
use super::*;

fn rendered_wire_bounds(
    wires: &[WireModel],
    fallback_z: f64,
) -> Option<(glam::DVec3, glam::DVec3)> {
    let mut min = glam::DVec3::splat(f64::INFINITY);
    let mut max = glam::DVec3::splat(f64::NEG_INFINITY);
    macro_rules! include_bound {
        ($p:expr) => {
            let point: glam::DVec3 = $p;
            if point.is_finite() {
                min = min.min(point);
                max = max.max(point);
            }
        };
    }

    for wire in wires {
        let mut has_precise_position = false;
        let hw = wire.world_width as f64 * 0.5;

        for (index, &[x, y, z]) in wire.points.iter().enumerate() {
            let low = wire.points_low.get(index).copied().unwrap_or([0.0; 3]);
            let p = glam::DVec3::new(
                x as f64 + low[0] as f64,
                y as f64 + low[1] as f64,
                z as f64 + low[2] as f64,
            );
            if hw > 0.0 {
                include_bound!(p - glam::DVec3::splat(hw));
                include_bound!(p + glam::DVec3::splat(hw));
            } else {
                include_bound!(p);
            }
            has_precise_position = true;
        }
        for vertex in &wire.text_verts {
            include_bound!(glam::DVec3::new(
                vertex.pos[0] as f64 + vertex.pos_low[0] as f64,
                vertex.pos[1] as f64 + vertex.pos_low[1] as f64,
                vertex.pos[2] as f64 + vertex.pos_low[2] as f64,
            ));
            has_precise_position = true;
        }
        for &[x, y, z] in &wire.key_vertices {
            let p = glam::DVec3::new(x, y, z);
            if hw > 0.0 {
                include_bound!(p - glam::DVec3::splat(hw));
                include_bound!(p + glam::DVec3::splat(hw));
            } else {
                include_bound!(p);
            }
            has_precise_position = true;
        }
        for &(point, _) in &wire.snap_pts {
            include_bound!(point);
            has_precise_position = true;
        }

        if !has_precise_position {
            let [x0, y0, x1, y1] = wire.aabb;
            if wire.aabb != WireModel::UNBOUNDED_AABB
                && [x0, y0, x1, y1].iter().all(|value| value.is_finite())
                && x0 <= x1
                && y0 <= y1
            {
                include_bound!(glam::DVec3::new(
                    x0 as f64 - hw,
                    y0 as f64 - hw,
                    fallback_z - hw
                ));
                include_bound!(glam::DVec3::new(
                    x1 as f64 + hw,
                    y1 as f64 + hw,
                    fallback_z + hw
                ));
            }
        }
    }

    if min.is_finite() && max.is_finite() {
        Some((min, max))
    } else {
        None
    }
}

/// Merge `pair` into `bounds`, dropping coordinates that are parser junk
/// rather than geometry: DXF writes ±1e20 as its "no extents" sentinel, and
/// past 1e16 an f32 has no resolution left to place a point with.
fn fold_sane_bounds(
    bounds: &mut Option<(glam::DVec3, glam::DVec3)>,
    pair: Option<(glam::DVec3, glam::DVec3)>,
) {
    const SANE_EXTENT: f64 = 1.0e16;
    let Some((lo, hi)) = pair else {
        return;
    };
    if lo.abs().max_element() >= SANE_EXTENT || hi.abs().max_element() >= SANE_EXTENT {
        return;
    }
    *bounds = Some(match *bounds {
        Some((min, max)) => (min.min(lo), max.max(hi)),
        None => (lo, hi),
    });
}

fn rendered_wire_center(wires: &[WireModel], fallback_z: f64) -> Option<glam::DVec3> {
    rendered_wire_bounds(wires, fallback_z).map(|(min, max)| (min + max) * 0.5)
}

fn block_entity_transform(
    document: &CadDocument,
    block_name: &str,
    target: Handle,
    visited: &mut Vec<String>,
    annotation_scale: f32,
) -> Option<codec::types::Transform> {
    if visited
        .iter()
        .any(|name| name.eq_ignore_ascii_case(block_name))
    {
        return None;
    }
    let record = document
        .block_records
        .iter()
        .find(|record| record.name.eq_ignore_ascii_case(block_name))?;
    if record.entity_handles.contains(&target) {
        return Some(codec::types::Transform::identity());
    }

    visited.push(record.name.clone());
    let transform = record.entity_handles.iter().find_map(|handle| {
        let entity = document.get_entity(*handle)?;
        crate::scene::render_graph::entity_render_block_uses(document, entity, annotation_scale)
            .into_iter()
            .filter(|block_use| block_use.active)
            .find_map(|block_use| {
                let inner = block_entity_transform(
                    document,
                    &block_use.insert.block_name,
                    target,
                    visited,
                    annotation_scale,
                )?;
                Some(
                    inner.then(&crate::scene::render_graph::insert_transform_with_policy(
                        document,
                        &block_use.insert,
                        annotation_scale,
                        block_use.scale_policy,
                    )),
                )
            })
    });
    visited.pop();
    transform
}

/// The model-space active viewport is reserved as `*Active`, but that name is
/// case-insensitive in DXF/DWG — a file may store it as `*ACTIVE`. Match it
/// accordingly, otherwise an uppercased record reads as a *distinct* viewport:
/// on save it survives the "drop the old *Active" filter and sits next to the
/// fresh entry we add, so the file gets two overlapping full-window viewports
/// that render "one on top of the other". (#319)
fn is_active_vport_name(name: &str) -> bool {
    name.eq_ignore_ascii_case("*Active")
}

/// The world point a named view is centred on. The VIEW record's centre is
/// in view (DCS) coordinates, an offset from its target in the view plane.
/// Views saved by earlier versions of this application stored the target
/// itself as the centre; that centre (equal to a non-zero target) is taken
/// as the target, not as an offset from it.
pub fn named_view_center(view: &codec::tables::View) -> glam::DVec3 {
    use glam::{DQuat, DVec3};
    let target = DVec3::new(view.target.x, view.target.y, view.target.z);
    let (cx, cy) = (view.center.x, view.center.y);
    let tol = 1e-9 * target.length().max(1.0);
    let legacy = (target.x != 0.0 || target.y != 0.0)
        && (cx - target.x).abs() <= tol
        && (cy - target.y).abs() <= tol;
    if legacy {
        return target;
    }
    let eye = DVec3::new(view.direction.x, view.direction.y, view.direction.z)
        .try_normalize()
        .unwrap_or(DVec3::Z);
    target + DQuat::from_rotation_arc(DVec3::Z, eye) * DVec3::new(cx, cy, 0.0)
}

impl Scene {
    /// A geometry mutation makes every fitted Model camera AABB stale. Clear
    /// both the live camera and inactive tile snapshots immediately so no view
    /// can clip against the old drawing until its bounds are refreshed.
    pub(super) fn invalidate_projection_bounds(&self) {
        self.projection_bounds_epoch.set(0);
        self.camera.borrow_mut().invalidate_model_bounds();
        for tile in self.model_tiles.borrow_mut().iter_mut() {
            tile.camera.invalidate_model_bounds();
        }
    }

    // ── Hit-test convenience: wire name → Handle ──────────────────────────

    pub fn handle_from_wire_name(name: &str) -> Option<Handle> {
        crate::scene::pipeline::wire_gpu::fast_parse_u64(name).map(Handle::new)
    }

    /// Restore camera to a named view from the document view table.
    pub fn restore_named_view(&mut self, view: &codec::tables::View) {
        use glam::Vec3;
        // The model viewport takes a copy of the view's background, or none.
        if self.current_layout == "Model" {
            let background = copy_background(&mut self.document, view.background_handle);
            self.set_active_background(background);
        }
        let cam = &mut *self.camera.borrow_mut();
        // The stored direction points from the target toward the eye.
        let eye_dir = Vec3::new(
            view.direction.x as f32,
            view.direction.y as f32,
            view.direction.z as f32,
        );
        let eye_dir = if eye_dir.length_squared() > 1e-10 {
            eye_dir.normalize()
        } else {
            Vec3::Z
        };
        // Build rotation: canonical eye is +Z, rotate to eye_dir.
        cam.rotation = glam::Quat::from_rotation_arc(Vec3::Z, eye_dir);
        cam.target = named_view_center(view);
        // Sync yaw/pitch from new rotation (for ViewCube).
        let pitch = eye_dir.z.clamp(-1.0, 1.0).asin();
        let yaw = eye_dir.x.atan2(eye_dir.y);
        cam.yaw = yaw;
        cam.pitch = pitch;
        // Derive distance from view height and fov.
        let h = view.height as f32;
        cam.distance = if h > 0.0 {
            h / (2.0 * (cam.fov_y * 0.5).tan())
        } else {
            cam.distance
        };
        self.camera_generation += 1;
    }

    /// The background the active model-space viewport shows (null for the
    /// drawing area colour).
    pub fn active_background(&self) -> Handle {
        self.active_tile_vport_handle()
            .and_then(|handle| self.document.vports.iter().find(|value| value.handle == handle))
            .map_or(Handle::NULL, |value| value.background_handle)
    }

    fn set_active_background(&mut self, background: Handle) {
        let Some(handle) = self.active_tile_vport_handle() else { return };
        if let Some(value) = self.document.vports.iter_mut().find(|value| value.handle == handle) {
            value.background_handle = background;
            self.camera_generation += 1;
        }
    }

    /// Save the current camera state into a new named view entry.
    /// Returns the view; caller must push it into document.views.
    pub fn current_as_named_view(&self, name: &str) -> codec::tables::View {
        use codec::types::Vector3;
        let cam = self.camera.borrow();
        let eye_dir = cam.rotation * glam::Vec3::Z;
        let height = cam.ortho_size() * 2.0;
        // The window's shape: the last rendered aspect ratio.
        let width = height * self.active_camera_aspect();
        let mut view = codec::tables::View::new(name);
        // The centre is in view (DCS) coordinates, relative to the target.
        view.center = Vector3 {
            x: 0.0,
            y: 0.0,
            z: 0.0,
        };
        view.target = Vector3 {
            x: cam.target.x as f64,
            y: cam.target.y as f64,
            z: cam.target.z as f64,
        };
        view.direction = Vector3 {
            x: eye_dir.x as f64,
            y: eye_dir.y as f64,
            z: eye_dir.z as f64,
        };
        view.height = height as f64;
        view.width = width as f64;
        view.lens_length = 50.0;
        view.front_clip = 0.0;
        view.back_clip = 0.0;
        view.twist_angle = 0.0;
        // OCS saves an orthographic model view, not a perspective camera.
        view.perspective = false;
        view
    }

    /// Zoom the model-space camera in/out by a percentage.
    /// factor > 1 = zoom out, factor < 1 = zoom in.
    pub fn zoom_camera(&mut self, factor: f32) {
        let mut cam = self.camera.borrow_mut();
        cam.distance = (cam.distance * factor).max(0.001);
        drop(cam);
        self.camera_generation += 1;
    }

    /// Save the active view immediately before a navigation operation. The
    /// paper-space camera and an entered floating viewport use different state,
    /// so the snapshot records whichever one currently owns navigation.
    pub fn remember_current_view(&mut self) {
        let layout = self.current_layout.clone();
        if let Some(handle) = self.active_viewport {
            if let Some(EntityType::Viewport(viewport)) = self.document.get_entity(handle) {
                self.previous_view = Some(ViewSnapshot::Floating {
                    layout,
                    handle,
                    viewport: Box::new(viewport.clone()),
                });
                return;
            }
        }
        self.previous_view = Some(ViewSnapshot::Main {
            layout,
            tile: self.active_model_tile.get(),
            camera: self.camera.borrow().clone(),
        });
    }

    /// Swap the active view with the most recently remembered one. Keeping the
    /// displaced current view makes repeated ZOOM Previous calls toggle cleanly
    /// between the last two views.
    pub fn restore_previous_view(&mut self) -> bool {
        let Some(snapshot) = self.previous_view.take() else {
            return false;
        };
        match snapshot {
            ViewSnapshot::Main {
                layout,
                tile,
                camera,
            } if self.active_viewport.is_none()
                && self.current_layout == layout
                && (layout != "Model" || self.active_model_tile.get() == tile) =>
            {
                let current = ViewSnapshot::Main {
                    layout,
                    tile,
                    camera: self.camera.borrow().clone(),
                };
                *self.camera.borrow_mut() = camera;
                self.previous_view = Some(current);
                self.camera_generation += 1;
                true
            }
            ViewSnapshot::Floating {
                layout,
                handle,
                viewport,
            } if self.current_layout == layout && self.active_viewport == Some(handle) => {
                let Some(EntityType::Viewport(current)) = self.document.get_entity(handle) else {
                    self.previous_view = Some(ViewSnapshot::Floating {
                        layout,
                        handle,
                        viewport,
                    });
                    return false;
                };
                if current.status.locked {
                    self.previous_view = Some(ViewSnapshot::Floating {
                        layout,
                        handle,
                        viewport,
                    });
                    return false;
                }
                let current = Box::new(current.clone());
                if let Some(EntityType::Viewport(destination)) =
                    self.document.get_entity_mut(handle)
                {
                    destination.view_center = viewport.view_center;
                    destination.view_direction = viewport.view_direction;
                    destination.view_target = viewport.view_target;
                    destination.lens_length = viewport.lens_length;
                    destination.front_clip_z = viewport.front_clip_z;
                    destination.back_clip_z = viewport.back_clip_z;
                    destination.view_height = viewport.view_height;
                    destination.twist_angle = viewport.twist_angle;
                    destination.custom_scale = viewport.custom_scale;
                    destination.status.perspective = viewport.status.perspective;
                }
                self.previous_view = Some(ViewSnapshot::Floating {
                    layout,
                    handle,
                    viewport: current,
                });
                self.camera_generation += 1;
                true
            }
            snapshot => {
                self.previous_view = Some(snapshot);
                false
            }
        }
    }

    /// Refresh model bounds without moving the live camera. Reusing `fit_all`
    /// keeps projection framing on the same visibility and outlier filters as
    /// Zoom Extents instead of accepting every finite cached AABB.
    pub(crate) fn refresh_projection_bounds(&mut self) {
        if self.current_layout != "Model"
            || self.active_viewport.is_some()
            || (self.projection_bounds_epoch.get() == self.geometry_epoch
                && self.camera.borrow().fitted_model_bounds().is_some())
        {
            return;
        }

        let saved_camera = self.camera.borrow().clone();
        let saved_generation = self.camera_generation;
        self.camera.borrow_mut().invalidate_model_bounds();
        self.fit_all();
        let refreshed = self.camera.borrow().fitted_model_bounds();
        *self.camera.borrow_mut() = saved_camera;
        self.camera_generation = saved_generation;

        if let Some((min, max)) = refreshed {
            self.camera.borrow_mut().fit_depth_to_bounds_f64(min, max);
            self.projection_bounds_epoch.set(self.geometry_epoch);
        }
    }

    /// Switch the live camera projection without changing its target or
    /// orientation. Model layout bounds are refreshed first so perspective eye
    /// distance accounts for the current, visible drawing depth.
    pub fn set_projection_preserving_frame(
        &mut self,
        projection: view::camera::Projection,
    ) -> bool {
        if self.camera.borrow().projection == projection {
            return false;
        }

        self.refresh_projection_bounds();

        let aspect = self.active_camera_aspect();
        self.camera
            .borrow_mut()
            .set_projection_preserving_frame(projection, aspect);
        self.camera_generation += 1;
        true
    }

    /// Aspect ratio of the active camera's actual render rectangle. Model
    /// panes are separate shader widgets, so the full-canvas render aspect is
    /// wrong whenever a tiled viewport is active.
    pub(super) fn active_camera_aspect(&self) -> f32 {
        if self.current_layout == "Model" {
            let (canvas_w, canvas_h) = self.selection.borrow().view.vp_size;
            let tiles = self.model_tiles.borrow();
            let active = self
                .active_model_tile
                .get()
                .min(tiles.len().saturating_sub(1));
            if let Some(tile) = tiles.get(active) {
                let width = tile.rect.width * canvas_w;
                let height = tile.rect.height * canvas_h;
                if width.is_finite() && height.is_finite() && width > 0.0 && height > 0.0 {
                    return width / height;
                }
            }
        }
        self.last_render_aspect.get().max(0.01)
    }

    /// Fit the camera to a world-space bounding box (corners p1, p2).
    pub fn zoom_to_window(&mut self, p1: glam::Vec3, p2: glam::Vec3) {
        let min = p1.min(p2);
        let max = p1.max(p2);
        if min == max {
            return;
        }
        let aspect = self.active_camera_aspect();
        self.camera.borrow_mut().fit_to_bounds(min, max, aspect);
        self.projection_bounds_epoch.set(self.geometry_epoch);
        self.camera_generation += 1;
    }

    /// Fit the active view to the rendered bounds of the supplied entities.
    /// Returns false when the selection has no finite visible geometry.
    pub fn zoom_to_entities(&mut self, handles: &[Handle]) -> bool {
        let fallback_z = self.active_camera_target().z;
        let wires = self.wire_models_for(handles);
        let mut bounds = rendered_wire_bounds(&wires, fallback_z);

        for handle in handles {
            let Some(mesh) = self.meshes.get(handle) else {
                continue;
            };
            let [x0, y0, x1, y1] = mesh.world_aabb;
            let [z0, z1] = mesh.z_aabb;
            let mesh_min = glam::DVec3::new(x0 as f64, y0 as f64, z0 as f64);
            let mesh_max = glam::DVec3::new(x1 as f64, y1 as f64, z1 as f64);
            if !mesh_min.is_finite() || !mesh_max.is_finite() {
                continue;
            }
            bounds = Some(match bounds {
                Some((min, max)) => (min.min(mesh_min), max.max(mesh_max)),
                None => (mesh_min, mesh_max),
            });
        }

        let Some((mut min, mut max)) = bounds else {
            return false;
        };
        if min == max {
            min -= glam::DVec3::splat(1.0);
            max += glam::DVec3::splat(1.0);
        }
        let min = min.as_vec3();
        let max = max.as_vec3();
        if !min.is_finite() || !max.is_finite() {
            return false;
        }
        if self.active_viewport.is_some() {
            return self.fit_active_viewport_to_bounds(min, max);
        }
        let aspect = self.active_camera_aspect();
        self.camera.borrow_mut().fit_to_bounds(min, max, aspect);
        self.projection_bounds_epoch.set(self.geometry_epoch);
        self.camera_generation += 1;
        true
    }

    /// Centre the active camera on one rendered entity without changing zoom
    /// distance, orientation, or viewport scale.
    pub fn center_camera_on_entity(&mut self, handle: Handle) -> bool {
        let fallback_z = self.active_camera_target().z;
        let wires = self.wire_models_for(&[handle]);
        let Some(center) = rendered_wire_center(&wires, fallback_z) else {
            return false;
        };
        self.center_active_camera_on(center)
    }

    /// Centre on one concrete attribute value attached to an INSERT rather
    /// than on the complete block extents.
    pub fn center_camera_on_insert_attribute(&mut self, insert: Handle, index: usize) -> bool {
        let (attribute, fallback) = {
            let Some(EntityType::Insert(entity)) = self.document.get_entity(insert) else {
                return false;
            };
            let Some(attribute) = entity.attributes.get(index) else {
                return false;
            };
            (
                EntityType::AttributeEntity(attribute.clone()),
                glam::DVec3::new(
                    attribute.insertion_point.x,
                    attribute.insertion_point.y,
                    attribute.insertion_point.z,
                ),
            )
        };
        let wires = self.tessellate_one(&attribute);
        let center = rendered_wire_center(&wires, fallback.z).unwrap_or(fallback);
        self.center_active_camera_on(center)
    }

    /// Centre on a text entity stored in a block definition, transformed
    /// through the concrete visible INSERT occurrence that FIND is visiting.
    pub fn center_camera_on_block_entity(&mut self, insert: Handle, entity: Handle) -> bool {
        let (source, transform) = {
            let Some(EntityType::Insert(insert_entity)) = self.document.get_entity(insert) else {
                return false;
            };
            let Some(source) = self.document.get_entity(entity) else {
                return false;
            };
            let mut visited = Vec::new();
            let Some(inner) = block_entity_transform(
                &self.document,
                &insert_entity.block_name,
                entity,
                &mut visited,
                self.annotation_scale,
            ) else {
                return false;
            };
            (
                source.clone(),
                inner.then(&crate::scene::render_graph::insert_transform_at_scale(
                    &self.document,
                    insert_entity,
                    self.annotation_scale,
                )),
            )
        };

        let fallback_z = self.active_camera_target().z;
        let wires = self.tessellate_one(&source);
        let Some(local_center) = rendered_wire_center(&wires, fallback_z) else {
            return false;
        };
        let world = transform.apply(codec::types::Vector3::new(
            local_center.x,
            local_center.y,
            local_center.z,
        ));
        self.center_active_camera_on(glam::DVec3::new(world.x, world.y, world.z))
    }

    fn active_camera_target(&self) -> glam::DVec3 {
        self.active_viewport
            .and_then(|handle| self.camera_for_viewport(handle))
            .map_or_else(|| self.camera.borrow().target, |camera| camera.target)
    }

    fn center_active_camera_on(&mut self, center: glam::DVec3) -> bool {
        if !center.is_finite() {
            return false;
        }
        if let Some(handle) = self.active_viewport {
            let Some(EntityType::Viewport(viewport)) = self.document.get_entity_mut(handle) else {
                return false;
            };
            if viewport.status.locked {
                return false;
            }
            viewport.view_target.x = center.x;
            viewport.view_target.y = center.y;
            viewport.view_target.z = center.z;
            self.notify_viewport_changed(handle);
        } else {
            self.camera.borrow_mut().target = center;
        }
        self.camera_generation += 1;
        true
    }

    /// Apply camera state from an opencadcodec View table entry, through the shared
    /// `camera_from_view` decoder so the twist round-trips like every other
    /// saved view. `model_space`: if true, subtracts world_offset from target
    /// (wire-space); paper-space entries carry no offset.
    fn apply_camera_from_view_entry(
        &mut self,
        view: &codec::tables::View,
        model_space: bool,
    ) -> bool {
        let _ = model_space;
        let Some(cam) = self.camera_from_view_mode(
            view.direction,
            view.target,
            codec::types::Vector2 {
                x: view.center.x,
                y: view.center.y,
            },
            view.height,
            view.twist_angle,
            view.perspective,
            view.lens_length,
        ) else {
            return false;
        };
        *self.camera.borrow_mut() = cam;
        self.camera_generation += 1;
        true
    }

    /// Set the model-space camera from the VPORT table's *Active entry.
    /// Returns true if the entry was found and the camera was set.
    fn apply_active_vport_camera(&mut self) -> bool {
        // Restore the single tile's visual style + grid/snap from the *Active
        // entry, independent of where the camera itself comes from below.
        if let Some(vp) = self
            .document
            .vports
            .iter()
            .find(|v| is_active_vport_name(&v.name))
        {
            let mode = vp.render_mode;
            let (grid_on, snap_on) = (vp.grid_on, vp.snap_on);
            let mut tiles = self.model_tiles.borrow_mut();
            let active = self
                .active_model_tile
                .get()
                .min(tiles.len().saturating_sub(1));
            if let Some(t) = tiles.get_mut(active) {
                t.render_mode = mode;
                t.grid_on = grid_on;
                t.snap_on = snap_on;
            }
        }
        // Restore from the standard *Active VPORT entry. (Earlier builds also
        // wrote an app-specific "OpenCADStudio_Camera_Model" View record and
        // preferred it here — that polluted the file for other CAD programs and
        // is no longer written or read; the view round-trips fine via VPORT.)
        let vp = match self
            .document
            .vports
            .iter()
            .find(|v| is_active_vport_name(&v.name))
        {
            Some(v) => v.clone(),
            None => return false,
        };
        let Some(new_cam) = self.camera_from_vport(&vp) else {
            return false;
        };
        *self.camera.borrow_mut() = new_cam;
        self.camera_generation += 1;
        true
    }

    /// Decode a saved view into a `Camera`. This is the single shared decoder
    /// for both a model-space VPORT table entry (tiled) and a paper-space
    /// VIEWPORT entity (floating): tiled vs floating only changes *where* the
    /// fields come from and the floating auto-fit fallback — the projection
    /// math (view direction → yaw/pitch, twist → roll, view_center fold,
    /// view_height → distance) is identical, so it lives here once. Callers pass
    /// their already-effective `view_target` / `view_center` / `view_height`.
    ///
    /// Returns `None` for a zero `view_height` (an uninitialised entry).
    pub(super) fn camera_from_view_mode(
        &self,
        view_direction: codec::types::Vector3,
        view_target: codec::types::Vector3,
        view_center: codec::types::Vector2,
        view_height: f64,
        twist: f64,
        perspective: bool,
        lens_length: f64,
    ) -> Option<Camera> {
        if view_height.abs() < 1e-9 {
            return None;
        }
        let vd = glam::Vec3::new(
            view_direction.x as f32,
            view_direction.y as f32,
            view_direction.z as f32,
        )
        .normalize_or(glam::Vec3::Z);
        let pitch = vd.z.clamp(-1.0, 1.0).asin();
        // view_dir = (sin(yaw)*cos(pitch), -cos(yaw)*cos(pitch), sin(pitch))
        // → yaw = atan2(x, -y), but when looking straight up/down cos(pitch)≈0
        //   both x and y are near zero and atan2(0, -0.0) = π due to IEEE 754.
        let yaw = if vd.x.abs() < 1e-6 && vd.y.abs() < 1e-6 {
            0.0_f32 // plan/nadir view: yaw is undefined, default to 0
        } else {
            vd.x.atan2(-vd.y)
        };
        // The saved view can carry a twist (rotation about the view axis), set
        // when the view was aligned to a rotated UCS. The twist is the angle
        // that rotates world-X onto screen-right, so the world direction that
        // ends up horizontal is its negative; feed that in as the camera roll
        // so the drawing opens square, the way it was saved, instead of in raw
        // world orientation (which looks tilted).
        let rotation = view::camera::yaw_pitch_to_quat(yaw, pitch, -twist as f32);
        let view_right = rotation * glam::Vec3::X;
        let view_up = rotation * glam::Vec3::Y;
        // view_target is WCS; wire-space subtracts world_offset. view_center is
        // a DCS (screen-plane) offset, so fold it through the view basis.
        // Keep the target in f64: casting it to f32 first quantizes the camera
        // to ~0.5 m at UTM scale, so panning/zooming inside a floating viewport
        // (which nudges view_target by sub-metre f64 steps) made the content
        // jump on the f32 grid. The axis directions stay f32 (orientation only);
        // only the position must stay precise — matching the model camera.
        let base = glam::DVec3::new(view_target.x, view_target.y, view_target.z);
        let target =
            base + view_right.as_dvec3() * view_center.x + view_up.as_dvec3() * view_center.y;
        let lens_length = lens_length.max(1.0) as f32;
        // The saved-view convention uses a 24 mm vertical aperture. This is
        // the inverse of the projection path's documented
        // `distance = view_height * lens / 24` relation.
        let fov_y = 2.0 * (12.0 / lens_length).atan();
        let distance = ((view_height as f32 / 2.0) / (fov_y * 0.5).tan()).max(0.001);
        Some(Camera {
            target,
            rotation,
            distance,
            fov_y,
            projection: if perspective {
                view::camera::Projection::Perspective
            } else {
                view::camera::Projection::Orthographic
            },
            yaw,
            pitch,
            // Sized from the saved view height (not a zoom-scaled range) so a
            // restored view keeps stable depth precision. A THIN projection —
            // e.g. a model-documentation section/side view whose cut plane sits
            // half an object-depth from the view target — has content much
            // deeper than its visible height, so the depth half-range must be a
            // generous multiple: `×4` clipped a section view's outline that was
            // ~6× the view height deep. Orthographic depth precision is linear,
            // so the wide range costs no z-fighting.
            depth_half_range: (view_height as f32 * 64.0).max(1.0),
            // A restored view has no model AABB yet; `fit_depth_to_saved_extents`
            // fills it in from $EXTMIN/$EXTMAX right after, so the near/far tracks
            // the eye direction once the view is orbited.
            model_bounds: None,
        })
    }

    /// Decode a VPort table entry (model-space tiled view) into a `Camera`.
    fn camera_from_vport(&self, vp: &codec::tables::VPort) -> Option<Camera> {
        self.camera_from_view_mode(
            vp.view_direction,
            vp.view_target,
            vp.view_center,
            vp.view_height,
            vp.view_twist,
            vp.perspective,
            vp.lens_length,
        )
    }

    fn apply_camera_to_vport(
        &self,
        entry: &mut codec::tables::VPort,
        cam: &Camera,
        lower_left: codec::types::Vector2,
        upper_right: codec::types::Vector2,
    ) {
        let view_dir = cam.rotation * glam::Vec3::Z;
        let view_height = cam.ortho_size() * 2.0;
        let target_wcs = codec::types::Vector3 {
            x: cam.target.x,
            y: cam.target.y,
            z: cam.target.z,
        };
        entry.lower_left = lower_left;
        entry.upper_right = upper_right;
        entry.view_target = target_wcs;
        entry.view_direction = codec::types::Vector3 {
            x: view_dir.x as f64,
            y: view_dir.y as f64,
            z: view_dir.z as f64,
        };
        entry.view_height = view_height as f64;
        let (canvas_width, canvas_height) = self.selection.borrow().view.vp_size;
        let viewport_width = (upper_right.x - lower_left.x).abs() * canvas_width as f64;
        let viewport_height = (upper_right.y - lower_left.y).abs() * canvas_height as f64;
        if viewport_width > 1e-9 && viewport_height > 1e-9 {
            entry.aspect_ratio = viewport_width / viewport_height;
        }
        entry.perspective = cam.projection == view::camera::Projection::Perspective;
        entry.lens_length = (12.0 / (cam.fov_y * 0.5).tan().max(1e-6)) as f64;
        entry.view_center = codec::types::Vector2::ZERO;
        // Stored twist = -roll, matching the decoder (roll = -twist).
        entry.view_twist = -cam.roll() as f64;
    }

    fn vport_rect_matches(
        entry: &codec::tables::VPort,
        lower_left: codec::types::Vector2,
        upper_right: codec::types::Vector2,
    ) -> bool {
        const EPSILON: f64 = 1e-5;
        (entry.lower_left.x - lower_left.x).abs() <= EPSILON
            && (entry.lower_left.y - lower_left.y).abs() <= EPSILON
            && (entry.upper_right.x - upper_right.x).abs() <= EPSILON
            && (entry.upper_right.y - upper_right.y).abs() <= EPSILON
    }

    fn vport_contains_rect_center(
        entry: &codec::tables::VPort,
        lower_left: codec::types::Vector2,
        upper_right: codec::types::Vector2,
    ) -> bool {
        const EPSILON: f64 = 1e-5;
        let center_x = (lower_left.x + upper_right.x) * 0.5;
        let center_y = (lower_left.y + upper_right.y) * 0.5;
        center_x >= entry.lower_left.x - EPSILON
            && center_x <= entry.upper_right.x + EPSILON
            && center_y >= entry.lower_left.y - EPSILON
            && center_y <= entry.upper_right.y + EPSILON
    }

    /// Convert a `ModelTile`'s normalized iced rectangle (top-left origin) to
    /// the (lower_left, upper_right) pair the VPort table uses (bottom-left
    /// origin).
    fn tile_rect_to_vport(rect: iced::Rectangle) -> (codec::types::Vector2, codec::types::Vector2) {
        let lower_left = codec::types::Vector2 {
            x: rect.x as f64,
            y: (1.0 - rect.y - rect.height) as f64,
        };
        let upper_right = codec::types::Vector2 {
            x: (rect.x + rect.width) as f64,
            y: (1.0 - rect.y) as f64,
        };
        (lower_left, upper_right)
    }

    /// Inverse of `tile_rect_to_vport`.
    fn vport_to_tile_rect(
        lower_left: codec::types::Vector2,
        upper_right: codec::types::Vector2,
    ) -> iced::Rectangle {
        iced::Rectangle {
            x: lower_left.x as f32,
            y: (1.0 - upper_right.y) as f32,
            width: (upper_right.x - lower_left.x) as f32,
            height: (upper_right.y - lower_left.y) as f32,
        }
    }

    /// Restore model tiles from duplicate `*Active` VPort entries.
    /// Returns true on success — the caller skips `apply_active_vport_camera`
    /// in that case because the active tile's camera has already been loaded
    /// into `self.camera`.
    fn restore_model_tiles_from_vports(&mut self) -> bool {
        let active_vports: Vec<codec::tables::VPort> = self
            .document
            .vports
            .iter()
            .filter(|v| is_active_vport_name(&v.name))
            .cloned()
            .collect();

        if active_vports.len() <= 1 {
            return false;
        }

        let tiles: Vec<ModelTile> = active_vports
            .iter()
            .filter_map(|vp| {
                self.camera_from_vport(vp).map(|cam| ModelTile {
                    rect: Self::vport_to_tile_rect(vp.lower_left, vp.upper_right),
                    camera: cam,
                    render_mode: vp.render_mode,
                    grid_on: vp.grid_on,
                    snap_on: vp.snap_on,
                })
            })
            .collect();

        if tiles.len() <= 1 {
            return false;
        }

        let active_cam = tiles[0].camera.clone();
        *self.model_tiles.borrow_mut() = tiles;
        self.active_model_tile.set(0);
        *self.camera.borrow_mut() = active_cam;
        self.projection_bounds_epoch.set(0);
        // Rebuild the pane_grid layout from the restored tile rects so the panes
        // match (the renderer / input iterate panes, not tiles).
        self.rebuild_panes_from_tiles();
        self.camera_generation += 1;
        true
    }

    /// Persist model tiles as duplicate `*Active` VPort entries.
    fn save_model_tiles_to_vports(&mut self) {
        // Stash the live camera into the active tile so the about-to-write
        // snapshot reflects the user's most recent orbit / pan / zoom.
        {
            let live_cam = self.camera.borrow().clone();
            let mut tiles = self.model_tiles.borrow_mut();
            let active = self
                .active_model_tile
                .get()
                .min(tiles.len().saturating_sub(1));
            if let Some(t) = tiles.get_mut(active) {
                t.camera = live_cam;
            }
        }

        let table_handle = self.document.vports.handle();
        let mut active_vports: Vec<codec::tables::VPort> = self
            .document
            .vports
            .iter()
            .filter(|v| is_active_vport_name(&v.name))
            .cloned()
            .collect();
        let source_vports = active_vports.clone();
        let inherited_vport = active_vports.first().cloned();
        let preserved_vps: Vec<codec::tables::VPort> = self
            .document
            .vports
            .iter()
            .filter(|v| !is_active_vport_name(&v.name))
            .cloned()
            .collect();
        let mut new_vports = codec::tables::Table::with_handle(table_handle);
        for vp in preserved_vps {
            new_vports.add_or_replace(vp);
        }
        self.document.vports = new_vports;

        let tiles = self.model_tiles.borrow().clone();
        if tiles.is_empty() {
            return;
        }

        let active = self
            .active_model_tile
            .get()
            .min(tiles.len().saturating_sub(1));
        let mut ordered_tiles = Vec::with_capacity(tiles.len());
        ordered_tiles.push(tiles[active].clone());
        for (i, tile) in tiles.iter().enumerate() {
            if i != active {
                ordered_tiles.push(tile.clone());
            }
        }

        let mut scene_objects_changed = false;
        for tile in ordered_tiles {
            let (ll, ur) = Self::tile_rect_to_vport(tile.rect);
            let exact = active_vports
                .iter()
                .position(|entry| Self::vport_rect_matches(entry, ll, ur));
            let inherited = exact.or_else(|| {
                active_vports
                    .iter()
                    .enumerate()
                    .filter(|(_, entry)| Self::vport_contains_rect_center(entry, ll, ur))
                    .min_by(|(_, left), (_, right)| {
                        let left_area = (left.upper_right.x - left.lower_left.x)
                            * (left.upper_right.y - left.lower_left.y);
                        let right_area = (right.upper_right.x - right.lower_left.x)
                            * (right.upper_right.y - right.lower_left.y);
                        left_area.total_cmp(&right_area)
                    })
                    .map(|(index, _)| index)
            });
            let mut entry = inherited
                .map(|index| active_vports.remove(index))
                .or_else(|| {
                    source_vports
                        .iter()
                        .filter(|entry| Self::vport_contains_rect_center(entry, ll, ur))
                        .min_by(|left, right| {
                            let left_area = (left.upper_right.x - left.lower_left.x)
                                * (left.upper_right.y - left.lower_left.y);
                            let right_area = (right.upper_right.x - right.lower_left.x)
                                * (right.upper_right.y - right.lower_left.y);
                            left_area.total_cmp(&right_area)
                        })
                        .cloned()
                })
                .or_else(|| inherited_vport.clone())
                .unwrap_or_else(|| codec::tables::VPort::new("*Active"));
            let cloned = inherited.is_none();
            if cloned {
                entry.handle = codec::Handle::NULL;
            }
            entry.name = "*Active".to_string();
            self.apply_camera_to_vport(&mut entry, &tile.camera, ll, ur);
            entry.render_mode = tile.render_mode;
            // Each viewport persists its own grid display + grid-snap (#121).
            entry.grid_on = tile.grid_on;
            entry.snap_on = tile.snap_on;
            if !entry.handle.is_valid() {
                entry.handle = self.document.allocate_handle();
            }
            if cloned && entry.sun_handle.is_valid() {
                let source_sun = self
                    .document
                    .objects
                    .get(&entry.sun_handle)
                    .and_then(|object| match object {
                        codec::objects::ObjectType::ClassObject(value)
                            if matches!(&value.data, codec::objects::ClassObjectData::Sun(_)) =>
                        {
                            Some(value.clone())
                        }
                        _ => None,
                    });
                if let Some(mut sun) = source_sun {
                    let handle = self.document.allocate_handle();
                    sun.handle = handle;
                    sun.owner = entry.handle;
                    sun.reactors.clear();
                    sun.xdictionary_handle = None;
                    self.document
                        .objects
                        .insert(handle, codec::objects::ObjectType::ClassObject(sun));
                    entry.sun_handle = handle;
                    scene_objects_changed = true;
                }
            }
            self.document.vports.add_allow_duplicate(entry);
        }
        if scene_objects_changed {
            self.object_data_cache = crate::entities::object_data::build_cache(&self.document);
            self.lighting_cache.borrow_mut().clear();
        }
    }

    /// Mirror the live grid/snap toggles onto the active view's own store so the
    /// state stays independent per viewport: a model tile in model space, the
    /// layout's sheet viewport in paper space. (#121)
    pub fn set_active_tile_grid_snap(&mut self, grid_on: bool, snap_on: bool) {
        if self.current_layout != "Model" {
            // Paper space: target the active floating viewport if the user is
            // working inside one, otherwise the layout's sheet viewport. Each
            // viewport keeps its own grid/snap (round-tripped via status flags).
            let h = self
                .active_viewport
                .filter(|h| h.is_valid())
                .unwrap_or_else(|| self.current_layout_sheet_viewport_handle());
            if h.is_valid() {
                if let Some(EntityType::Viewport(vp)) = self.document.get_entity_mut(h) {
                    vp.status.grid_on = grid_on;
                    vp.status.snap_on = snap_on;
                }
            }
            return;
        }
        let mut tiles = self.model_tiles.borrow_mut();
        let active = self
            .active_model_tile
            .get()
            .min(tiles.len().saturating_sub(1));
        if let Some(t) = tiles.get_mut(active) {
            t.grid_on = grid_on;
            t.snap_on = snap_on;
        }
    }

    /// The active view's grid display + grid-snap, adopted into the live toggles
    /// on load and whenever the active viewport / tab / layout changes. Reads the
    /// model tile in model space, the sheet viewport in paper space. (#121)
    pub fn active_tile_grid_snap(&self) -> Option<(bool, bool)> {
        if self.current_layout != "Model" {
            let h = self
                .active_viewport
                .filter(|h| h.is_valid())
                .unwrap_or_else(|| self.current_layout_sheet_viewport_handle());
            if h.is_valid() {
                if let Some(EntityType::Viewport(vp)) = self.document.get_entity(h) {
                    return Some((vp.status.grid_on, vp.status.snap_on));
                }
            }
            return Some((false, false));
        }
        let tiles = self.model_tiles.borrow();
        let active = self
            .active_model_tile
            .get()
            .min(tiles.len().saturating_sub(1));
        tiles.get(active).map(|t| (t.grid_on, t.snap_on))
    }

    /// Set the paper-space camera from the sheet viewport's stored view.
    /// Returns true if a valid sheet viewport was found and the camera was set.
    ///
    /// The sheet viewport entity is the authoritative paper-space view (it
    /// round-trips through both the DXF and DWG writers). An older
    /// `OpenCADStudio_Camera_<layout>` named View is honoured only as a
    /// backward-compatible fallback for files saved under the previous scheme.
    fn apply_sheet_viewport_camera(&mut self) -> bool {
        // The Layout object already owns the authoritative sheet-viewport
        // handle. Do not scan every drawing entity on each Model↔Paper switch.
        let sheet_vp = match self
            .document
            .get_entity(self.current_layout_sheet_viewport_handle())
        {
            Some(EntityType::Viewport(vp)) => Some(vp.clone()),
            _ => None,
        };

        let vp = match sheet_vp {
            Some(v) if v.view_height.abs() >= 1e-9 => v,
            _ => {
                // Back-compat: files OCS saved with the named-View side-channel.
                let view_name = format!("OpenCADStudio_Camera_{}", self.current_layout);
                let fallback = self
                    .document
                    .views
                    .iter()
                    .find(|v| v.name == view_name)
                    .cloned();
                if let Some(view) = fallback {
                    return self.apply_camera_from_view_entry(&view, false);
                }
                return false;
            }
        };

        // Paper-space entities carry no world_offset → decode with a zero
        // offset, through the same shared decoder (twist included).
        let Some(cam) = self.camera_from_view_mode(
            vp.view_direction,
            vp.view_target,
            codec::types::Vector2 {
                x: vp.view_center.x,
                y: vp.view_center.y,
            },
            vp.view_height,
            vp.twist_angle,
            vp.status.perspective,
            vp.lens_length,
        ) else {
            return false;
        };
        *self.camera.borrow_mut() = cam;
        self.camera_generation += 1;
        true
    }

    /// Write the current camera back into the document (VPort or sheet viewport)
    /// so it is saved with the file. Returns true if the document was modified.
    pub fn sync_camera_to_document(&mut self) -> bool {
        let cam = self.camera.borrow().clone();
        let view_dir = cam.rotation * glam::Vec3::Z;
        let view_height = cam.ortho_size() * 2.0;
        // Stored twist is the negative of the camera roll (the decoder applies
        // roll = -twist), so the saved view round-trips square.
        let twist = -cam.roll() as f64;
        let vd3 = codec::types::Vector3 {
            x: view_dir.x as f64,
            y: view_dir.y as f64,
            z: view_dir.z as f64,
        };

        if self.current_layout == "Model" {
            self.save_model_tiles_to_vports();
            true
        } else {
            let target_wcs = codec::types::Vector3 {
                x: cam.target.x as f64,
                y: cam.target.y as f64,
                z: cam.target.z as f64,
            };

            // The sheet viewport entity is the authoritative paper-space view;
            // it round-trips natively, so no named-View side-channel is needed.
            let sheet_handle = self.current_layout_sheet_viewport_handle();
            if let Some(EntityType::Viewport(vp)) = self.document.get_entity_mut(sheet_handle) {
                // Paper-space position is stored in view_center (DCS).
                vp.view_center = codec::types::Vector3::new(target_wcs.x, target_wcs.y, 0.0);
                vp.view_target = codec::types::Vector3::ZERO;
                vp.view_direction = vd3;
                vp.view_height = view_height as f64;
                vp.twist_angle = twist;
                vp.status.perspective = cam.projection == view::camera::Projection::Perspective;
                vp.lens_length = (12.0 / (cam.fov_y * 0.5).tan().max(1e-6)) as f64;
            }
            true
        }
    }

    /// Restore the camera from the file's saved view (called once on open).
    /// Falls back to fit_all() if no saved view is available.
    pub fn restore_saved_camera(&mut self) {
        let restored = if self.current_layout == "Model" {
            // Tiled-layout restore takes precedence — it sets the camera too.
            // Single-tile files fall through to the *Active branch.
            self.restore_model_tiles_from_vports() || self.apply_active_vport_camera()
        } else {
            // Every paper layout has a full-screen sheet viewport that holds
            // its view; create one if a loaded file lacks it.
            let layout = self.current_layout.clone();
            self.ensure_sheet_viewport(&layout);
            self.apply_sheet_viewport_camera()
        };
        if restored {
            // The pose is the file's and must not move — but nothing has sized
            // its near/far, so `depth_half_range` stays unset and the
            // projection falls back to a guess scaled from the zoom distance.
            // On a drawing whose geometry reaches far off its plane that guess
            // is far too narrow and the geometry is silently clipped away.
            self.fit_depth_to_saved_extents();
        } else {
            self.fit_all();
        }
    }

    /// Size the camera's near/far from the drawing's own recorded extent,
    /// leaving its pose alone.
    ///
    /// `$EXTMIN`/`$EXTMAX` is untrustworthy for *positioning* — stale after an
    /// edit, or the 1e20 sentinel when the writer never computed it (see the
    /// `world_offset` selection notes) — which is why `fit_all` scans entities
    /// instead. It is the right input *here*: a near/far that is too wide only
    /// costs a little depth precision, while one that is too narrow deletes
    /// geometry outright, so erring on the drawing's own claim is the safe
    /// direction. It also costs nothing — no tessellation pass on open.
    fn fit_depth_to_saved_extents(&mut self) {
        // Same bound `model_space_extents`'s header fallback uses; rejects the
        // sentinel and any garbage that would blow the range up.
        const SANE_EXTENT: f64 = 1.0e16;
        let h = &self.document.header;
        let (lo, hi) = (h.model_space_extents_min, h.model_space_extents_max);
        let ok = lo.x <= hi.x
            && lo.y <= hi.y
            && lo.z <= hi.z
            && [lo.x, lo.y, lo.z, hi.x, hi.y, hi.z]
                .iter()
                .all(|v| v.is_finite() && v.abs() < SANE_EXTENT);
        if !ok {
            return;
        }
        let min = glam::DVec3::new(lo.x, lo.y, lo.z);
        let max = glam::DVec3::new(hi.x, hi.y, hi.z);
        self.camera.borrow_mut().fit_depth_to_bounds_f64(min, max);
    }

    fn fit_paper_space_extents(&mut self) {
        let layout_block = self.current_layout_block_handle();
        let mut min = glam::Vec3::splat(f32::INFINITY);
        let mut max = glam::Vec3::splat(f32::NEG_INFINITY);
        {
            let mut include = |x: f64, y: f64, z: f64| {
                const SANE_EXTENT: f64 = 1.0e16;
                if x.is_finite()
                    && y.is_finite()
                    && z.is_finite()
                    && x.abs() < SANE_EXTENT
                    && y.abs() < SANE_EXTENT
                    && z.abs() < SANE_EXTENT
                {
                    let point = glam::Vec3::new(x as f32, y as f32, z as f32);
                    min = min.min(point);
                    max = max.max(point);
                }
            };

            // The physical sheet is always part of Paper Space extents, even
            // when the layout contains no entities (#539).
            if let Some(((x0, y0), (x1, y1))) = self.paper_limits() {
                include(x0, y0, 0.0);
                include(x1, y1, 0.0);
            }

            // Paper entities and viewport borders belong to the sheet. Model
            // content projected through those viewports deliberately does not.
            let scale = if self.current_layout == "Model" {
                crate::scene::annotative::scale_handle_by_name(
                    &self.document,
                    &self.document.header.current_annotation_scale,
                )
            } else {
                self.paper_annotation_scale_handle()
            };
            for wire in self.wires_for_block_culled(
                layout_block,
                None,
                None,
                None,
                None,
                scale,
                self.annotation_all_visible(),
                None,
            ) {
                let is_infinite = Self::handle_from_wire_name(&wire.name)
                    .and_then(|handle| self.document.get_entity(handle))
                    .is_some_and(|entity| {
                        matches!(entity, EntityType::XLine(_) | EntityType::Ray(_))
                    });
                if is_infinite {
                    for &[x, y, z] in &wire.key_vertices {
                        include(x, y, z);
                    }
                } else {
                    for (index, &[x, y, z]) in wire.points.iter().enumerate() {
                        let low = wire.points_low.get(index).copied().unwrap_or([0.0; 3]);
                        include(
                            x as f64 + low[0] as f64,
                            y as f64 + low[1] as f64,
                            z as f64 + low[2] as f64,
                        );
                    }
                }
            }

            // Hatches, wipeouts and raster images are GPU-only and may carry
            // no normal wire outline, so include their visible paper bounds.
            let (hatches, wipeouts, images) = self.paper_sheet_render_models();
            for hatch in hatches.iter().chain(wipeouts.iter()) {
                for &[x, y] in hatch.boundary.iter() {
                    include(
                        hatch.world_origin[0] + x as f64,
                        hatch.world_origin[1] + y as f64,
                        0.0,
                    );
                }
            }
            for image in images.iter() {
                for (corner, low) in image.corners.iter().zip(image.corners_low.iter()) {
                    include(
                        corner[0] as f64 + low[0] as f64,
                        corner[1] as f64 + low[1] as f64,
                        corner[2] as f64 + low[2] as f64,
                    );
                }
            }
        }

        if !min.is_finite() || !max.is_finite() {
            return;
        }
        if min == max {
            max += glam::Vec3::splat(1.0);
        }
        self.camera
            .borrow_mut()
            .fit_to_bounds(min, max, self.last_render_aspect.get().max(0.01));
        self.camera_generation += 1;
    }

    /// Absolute-WCS bounds of everything visible in `block` — the one extents
    /// walk behind ZOOM EXTENTS and the drawing-extents readers.
    ///
    /// Nothing is rejected here on a guess. An earlier version dropped any wire
    /// whose centroid fell outside the inter-quartile consensus of the others,
    /// to stop one junk wire from a bad parse poisoning the box. It also
    /// dropped legitimately remote geometry — an outlying detail, a survey
    /// point, a block placed far from the rest — so ZOOM EXTENTS quietly
    /// refused to reach it. `SANE_EXTENT` and the finite test are the only
    /// filters now; the ±1e6 display segments that motivated the old heuristic
    /// come from XLine / Ray, which are handled explicitly below.
    pub(super) fn visible_block_bounds(&self, block: Handle) -> Option<(glam::DVec3, glam::DVec3)> {
        if block.is_null() {
            return None;
        }
        let mut bounds: Option<(glam::DVec3, glam::DVec3)> = None;
        macro_rules! fold {
            ($pair:expr) => {
                fold_sane_bounds(&mut bounds, $pair)
            };
        }

        // Drawing extents must not depend on where the camera happens to be
        // looking: the result is cached per geometry epoch, and a floating
        // viewport's own camera is derived FROM the extents
        // (`camera_for_viewport`), so reading the camera here recurses. A wire
        // with no precise position falls back to the z = 0 plane.
        let fallback_z = 0.0;
        let scale = crate::scene::annotative::scale_handle_by_name(
            &self.document,
            &self.document.header.current_annotation_scale,
        );
        // The resident set is the full, un-culled, LOD-free tessellation the
        // block actually renders from. Culled input would fit only what is
        // already on screen, so each ZOOM EXTENTS would zoom out a little and
        // reveal more, converging on the true extent after several uses (#51).
        let wires = self.resident_wires_for(block, None, scale, None, None);
        // Infinite display segments do not contribute to drawing extents.
        // Keep their base points as a fallback for otherwise empty drawings.
        let mut infinite_base: Vec<glam::DVec3> = Vec::new();
        for wire in wires.iter() {
            let is_infinite = Self::handle_from_wire_name(&wire.name)
                .and_then(|handle| self.document.get_entity(handle))
                .is_some_and(|entity| matches!(entity, EntityType::XLine(_) | EntityType::Ray(_)));
            if is_infinite {
                infinite_base.extend(
                    wire.key_vertices
                        .iter()
                        .map(|&[x, y, z]| glam::DVec3::new(x, y, z))
                        .filter(|point| point.is_finite()),
                );
                continue;
            }
            fold!(rendered_wire_bounds(std::slice::from_ref(wire), fallback_z));
        }

        // 3D solids render as meshes, not wires: a drawing of only solids has
        // no wire to fit.
        for (&handle, set) in &self.meshes {
            let Some(entity) = self.document.get_entity(handle) else {
                continue;
            };
            if !self.mesh_entity_visible(handle)
                || !self.belongs_to_visible_block(handle, entity.common().owner_handle, block)
            {
                continue;
            }
            let [x0, y0, x1, y1] = set.world_aabb;
            let [z0, z1] = set.z_aabb;
            let lo = glam::DVec3::new(x0 as f64, y0 as f64, z0 as f64);
            let hi = glam::DVec3::new(x1 as f64, y1 as f64, z1 as f64);
            if lo.is_finite() && hi.is_finite() {
                fold!(Some((lo.min(hi), lo.max(hi))));
            }
        }

        // Raster images, OLE frames and underlays draw as GPU quads; with
        // their frames hidden they carry no wire outline at all.
        for (&handle, image) in &self.images {
            let Some(entity) = self.document.get_entity(handle) else {
                continue;
            };
            if !self.mesh_entity_visible(handle)
                || !self.belongs_to_visible_block(handle, entity.common().owner_handle, block)
            {
                continue;
            }
            for (corner, low) in image.corners.iter().zip(image.corners_low.iter()) {
                let point = glam::DVec3::new(
                    corner[0] as f64 + low[0] as f64,
                    corner[1] as f64 + low[1] as f64,
                    corner[2] as f64 + low[2] as f64,
                );
                if point.is_finite() {
                    fold!(Some((point, point)));
                }
            }
        }

        // HATCH tessellates to no wires at all (#131), so a drawing whose
        // outermost geometry is a fill — or which is nothing but fills — has
        // to pick its bounds up from the fill models themselves. The paper
        // sheet already does this; model space did not.
        // The renderer already holds the fill models for the block it is
        // drawing, keyed by geometry epoch — reuse that rather than exploding
        // every INSERT again on each ZOOM EXTENTS. Only a block that is not
        // the one on screen (model space while a BEDIT is open) has to build
        // its own set.
        let hatches = if block == self.content_render_block_handle() {
            // Tinted, matching the renderer's own call: the two then share one
            // cache entry. Asking for an untinted set would key the same slot
            // with a different selection signature and the two would evict
            // each other on every frame. Tint is colour; it moves no vertex.
            self.hatch_models_arc_for_view(true)
        } else {
            std::sync::Arc::new(self.synced_hatch_models(
                block,
                None,
                scale,
                self.annotation_all_visible(),
                None,
                false,
            ))
        };
        for hatch in hatches.iter() {
            for &[x, y] in hatch.boundary.iter() {
                // NaN-NaN separators between disconnected paths fall out on
                // the finite test — they must not be treated as vertices.
                let point = glam::DVec3::new(
                    hatch.world_origin[0] + x as f64,
                    hatch.world_origin[1] + y as f64,
                    0.0,
                );
                if point.is_finite() {
                    fold!(Some((point, point)));
                }
            }
        }

        // Drawing holds only infinite construction geometry: fall back to its
        // base points rather than reporting nothing.
        if bounds.is_none() {
            for point in infinite_base {
                fold!(Some((point, point)));
            }
        }
        bounds
    }

    pub fn fit_all(&mut self) {
        if self.active_viewport.is_some() {
            if let Some((mut min, mut max)) = self.model_space_extents() {
                if min == max {
                    max += glam::Vec3::splat(1.0);
                    min -= glam::Vec3::splat(1.0);
                }
                self.fit_active_viewport_to_bounds(min, max);
            }
            return;
        }
        if self.current_layout != "Model" {
            self.fit_paper_space_extents();
            return;
        }

        let Some((min, mut max)) = self.visible_block_bounds(self.current_layout_block_handle())
        else {
            // Nothing usable to fit: leave the camera where it is.
            return;
        };
        if min == max {
            max += glam::DVec3::splat(1.0);
        }
        let aspect = self.active_camera_aspect();
        // f64 all the way into the camera: a UTM-scale drawing loses whole
        // metres to an f32 round-trip, which shows up as a fit that sits
        // slightly off the geometry it was asked to frame.
        self.camera.borrow_mut().fit_to_bounds_f64(min, max, aspect);
        self.projection_bounds_epoch.set(self.geometry_epoch);
        self.camera_generation += 1;
    }

    /// Fit every tiled Model viewport independently. The drawing bounds are
    /// shared, but each tile keeps its own orientation and uses its own screen
    /// aspect when calculating the required zoom.
    pub fn fit_all_model_viewports(&mut self) {
        if self.current_layout != "Model" || self.model_tiles.borrow().len() <= 1 {
            self.fit_all();
            return;
        }
        let generation_before = self.camera_generation;
        self.fit_all();
        if self.camera_generation == generation_before {
            return;
        }
        let Some((min, max)) = self.camera.borrow().fitted_model_bounds() else {
            return;
        };

        let (canvas_w, canvas_h) = self.selection.borrow().view.vp_size;
        let fallback = self.last_render_aspect.get().max(0.01);
        let aspects: Vec<f32> = self
            .model_tiles
            .borrow()
            .iter()
            .map(|tile| {
                let width = tile.rect.width * canvas_w;
                let height = tile.rect.height * canvas_h;
                if width.is_finite() && height.is_finite() && width > 0.0 && height > 0.0 {
                    width / height
                } else {
                    fallback
                }
            })
            .collect();
        let active = self
            .active_model_tile
            .get()
            .min(aspects.len().saturating_sub(1));
        let live_camera = self.camera.borrow_mut();
        let mut tiles = self.model_tiles.borrow_mut();
        for (index, tile) in tiles.iter_mut().enumerate() {
            let aspect = aspects.get(index).copied().unwrap_or(fallback);
            if index == active {
                tile.camera = live_camera.clone();
            } else {
                tile.camera.fit_to_bounds_f64(min, max, aspect);
            }
        }
        if let Some(aspect) = aspects.get(active) {
            self.last_render_aspect.set(*aspect);
        }
        self.camera_generation += 1;
    }

    pub fn update(&mut self, _dt: Duration) {
        // The glyph atlas grew or re-baked since this scene last built its
        // caches (a CJK sheet set bakes thousands of tiles on first sight):
        // every quad laid out before that samples the wrong tile — scrambled
        // text in the affected blocks / viewports. One structural bump
        // re-lays-out all text against the settled atlas.
        let atlas_gen = crate::scene::text::sdf_atlas::generation();
        if self.last_atlas_generation.get() != atlas_gen {
            self.last_atlas_generation.set(atlas_gen);
            self.bump_geometry();
        }
    }
}

/// The drawing's ACAD_BACKGROUND dictionary, made when it has none.
fn background_dictionary(document: &mut codec::CadDocument) -> Option<Handle> {
    use codec::objects::{Dictionary, ObjectType};
    let root = document.header.named_objects_dict_handle;
    let Some(ObjectType::Dictionary(root_dictionary)) = document.objects.get(&root) else {
        return None;
    };
    if let Some(existing) = root_dictionary.get("ACAD_BACKGROUND") {
        if matches!(document.objects.get(&existing), Some(ObjectType::Dictionary(_))) {
            return Some(existing);
        }
    }
    let handle = document.allocate_handle();
    let mut dictionary = Dictionary::new();
    dictionary.handle = handle;
    dictionary.owner = root;
    dictionary.reactors.push(root);
    document.objects.insert(handle, ObjectType::Dictionary(dictionary));
    if let Some(ObjectType::Dictionary(root_dictionary)) = document.objects.get_mut(&root) {
        root_dictionary.add_entry("ACAD_BACKGROUND", handle);
    }
    Some(handle)
}

/// A new background object in ACAD_BACKGROUND, keyed by its own handle as
/// the reference keys it. Null when the drawing has no named-objects root.
pub fn add_background(
    document: &mut codec::CadDocument,
    data: codec::objects::ClassObjectData,
) -> Handle {
    use codec::objects::{ClassObject, ObjectType};
    let Some(dictionary) = background_dictionary(document) else {
        return Handle::NULL;
    };
    let handle = document.allocate_handle();
    let mut object = ClassObject::new(data);
    object.handle = handle;
    object.owner = dictionary;
    object.reactors = vec![dictionary];
    document.objects.insert(handle, ObjectType::ClassObject(object));
    if let Some(ObjectType::Dictionary(entries)) = document.objects.get_mut(&dictionary) {
        entries.add_entry(format!("{:X}", handle.value()), handle);
    }
    handle
}

/// A copy of a background object (the reference gives each view and
/// viewport its own); null for none.
pub fn copy_background(document: &mut codec::CadDocument, handle: Handle) -> Handle {
    use codec::objects::{ClassObjectData as D, ObjectType};
    match document.objects.get(&handle) {
        Some(ObjectType::ClassObject(object))
            if matches!(
                object.data,
                D::SolidBackground(_) | D::GradientBackground(_) | D::ImageBackground(_)
                    | D::GroundPlaneBackground(_) | D::SkyLightBackground(_) | D::IblBackground(_)
            ) =>
        {
            let data = object.data.clone();
            add_background(document, data)
        }
        _ => Handle::NULL,
    }
}